A machine learning apparatus (20) that learns a control command for a machine (14) by machine learning, comprising: a machine learning unit (21, 25, 55) that performs machine learning to output the control command; a simulator (22, 52) that has a Simulates an operation of the machine (14) based on the control command; anda first determination unit (23, 53) which determines the control command on the basis of a result of the execution of the simulation by the simulator (22, 52), the first determination unit (23, 53) determining that the control command is valid, and an input of the for executes the control command found to be valid in the machine (14) if there is no problem in the execution result of the simulation, and determines that the control command is invalid, and the input of the control command found to be invalid in the machine (14) stops when a There is a problem and learning is performed by providing a first result mark (41, 61) obtained from the first determining unit (23, 53) as training data to the machine learning unit (21, 25, 55).
展开▼